Reporting on the Recently Possible
Every quarter we profile a different near future technology, producing a report on its development and a prototype demonstrating its application. Paired with our advising services, our reports and prototypes enable you to make the most of the recently possible.
Previews of our reports and prototypes are shown below, as well as links to select public prototypes. For access to all of our reports and prototypes contact us about becoming a subscriber.
Interpretability, or the ability to explain why and how a system makes a decision, can help us improve models, satisfy regulations, and build better products. Black-box techniques, such as deep learning, have delivered breakthrough capabilities at the cost of interpretability. In this report we show how to make models interpretable without sacrificing their capabilities or accuracy.
Refractor shows how interpretability opens up new product possibilities for machine learning applications. It predicts churn probabilities for telecom customers and shows which customer attributes are contributing to those predictions.
Probabilistic programming makes Bayesian inference accessible. It allows you to build interpretable models that quantify uncertainty and make use of both data and domain knowledge. In this report we show how to use probabilistic programming to build tools and products for more effective decision making.
Loan Officer Simulator
In Loan Officer Simulator you use graphs and metrics generated through a probabilistic model of loan repayments to decide which loans to approve and at what interest rate to approve them.
Probabilistic Real Estate predicts future real estate prices across the New York City boroughs. It enables you to input your housing budget and shows you the probability of finding properties in that price range across different neighborhoods and future time periods.
Summarization, and algorithms that make text quantifiable, allow us to derive insights from large amounts of unstructured text data. In this report we show to unlock the value of that data for new products and applications.
Brief is a browser extension that lets you highlight the interesting parts of any article on the internet. It uses neural networks to score and highlight the most interesting sentences within an article. Articles can then be read in 'highlight' or 'skim' mode.
A public version Brief that shows the results of running it on a limited selection of articles.
Deep Learning: Image Analysis
Deep learning, or highly-connected neural networks, offers fascinating new capabilities for image analysis. Using deep learning, computers can now learn to identify objects in images. This report explores the history and current state of the field, predicts future developments, and explains how to apply deep learning today.
Fathom allows you to explore our Instagram data set through content categories identified using image recognition. You can get a detailed view of how the algorithm works on photo detail pages, and you can explore related images using a category hierarchy.
Pictograph uses deep learning to analyze Instagram photos and reveal a person's top interests in the form of a pictograph.
Probabilistic Methods for Realtime Streams
Realtime analysis is a challenge for modern data systems. The probabilistic methods explored in this report offer highly efficient models for extracting value from streams of data as they are generated. This orders-of-magnitude improvement in efficiency enables many new applications and modes of computation.
CliqueStream displays popular topics across Reddit as nodes in a force-directed graph where the connections are based on similarity of word use over all comments and tweets. You can also select groups of keywords and view their distribution across subreddits.
Natural Language Generation
Machines are beginning to speak our language. Through natural language generation, computers can take highly structured data and output human language narrative. This report explores machine systems for natural language generation.
Inspired by real estate sites, RoboRealtor is a natural language generation prototype that takes the attributes for a hypothetical apartment and generates listings for it. Thousands of listings are generated and then ranked according to their relevance.